shapeup-sdlc 1.6.3 → 3.0.0

package/kernel/reduce/graph.mjs

@@ -0,0 +1,437 @@
+// graph — the run graph: one typed, append-only edge list per feature.
+//
+// WHAT IT IS FOR. The harness has always been graph-shaped and never a graph: orders produce
+// results, results produce verdicts, verdicts cite T0 artifacts, scopes cover requirements, trials
+// supersede trials. All of it was real and all of it was scattered across markdown and JSON, so
+// "trace this verdict back to the objective" meant grepping four files, and a relaunch re-derived
+// the whole run state by walking directories. This file makes the same facts one queryable thing.
+//
+// TWO FAMILIES, DELIBERATELY NOT COLLAPSED (they answer different questions and change at different
+// rates):
+//   WORK LINEAGE — Run, Order, Result, Verdict, Trial. "What happened, and what produced it."
+//   DOMAIN       — Scope, UseCase, Requirement, Seam. "What exists, and how it relates."
+// An edge may cross the families (a Scope COVERS a Requirement; an Order is DERIVED_FROM a Scope);
+// a node never belongs to both.
+//
+// THE GRAPH IS DERIVED, NEVER AUTHORED. Every node here is projected from an artifact already on
+// disk, so the graph can be deleted and rebuilt byte-identically, and a run that predates it is
+// backfilled by the same code path that maintains it. That is what makes it safe to treat as the
+// read model: it cannot drift from the artifacts, because it has no independent existence.
+//
+// SINGLE WRITER. `harness reduce graph` is the only thing that appends. It lives under `reduce/`
+// rather than `lib/` to say so in the file layout — `probe` imports it read-only.
+//
+// APPEND-ONLY, and idempotent. Re-running over an unchanged tree appends nothing; re-running after
+// new artifacts land appends only what is new. A node that reappears with new attributes is
+// appended again and the LAST line wins on read, so the file is a log and the projection is a fold.
+
+import { existsSync, readdirSync, readFileSync, appendFileSync, mkdirSync } from "node:fs";
+import { join, dirname, basename, resolve } from "node:path";
+import { runArgs } from "../lib/argv.mjs";
+import {
+  localRoot, receipt as receiptPath, ordersDir, resultsDir, verdictsDir, trials as trialsPath,
+  gates as gatesPath, scopesDir, usecasesDir, requirements as requirementsPath, wiringMap as wiringMapPath,
+} from "../lib/paths.mjs";
+import { readAllContracts, readContract, ucId, SCOPE_CONTRACT, WIRING_MAP } from "../lib/contract.mjs";
+import { runIdFromReceipt } from "../lib/paths.mjs";
+
+/** The graph's home — one file per feature, beside the run trace it projects. */
+export const graphPath = (cwd, slug) => join(localRoot(cwd, slug), "graph.jsonl");
+
+/** Node types, by family. A type outside these sets is a bug, not an extension point. */
+export const WORK_NODES = ["Run", "Order", "Result", "Verdict", "Trial", "GateDecision"];
+export const DOMAIN_NODES = ["Scope", "UseCase", "Requirement", "Seam"];
+
+/** Edge types. Each names a direction that is meaningful to read backwards. */
+export const EDGES = ["PRODUCED", "EVALUATES", "SUPERSEDES", "COVERS", "DEPENDS_ON", "DERIVED_FROM"];
+
+/**
+ * Read the graph as a log and fold it into nodes and edges.
+ *
+ * @param {string} cwd - Project root.
+ * @param {string} slug - Feature slug.
+ * @returns {{nodes: Map<string,object>, edges: Map<string,object>, lines: number}} The folded graph;
+ *   `lines` counts what was read, so a caller can tell an empty graph from an absent one.
+ */
+export function readGraph(cwd, slug) {
+  const nodes = new Map(), edges = new Map();
+  const p = graphPath(cwd, slug);
+  let lines = 0;
+  if (!existsSync(p)) return { nodes, edges, lines };
+  for (const line of readFileSync(p, "utf8").split("\n")) {
+    if (!line.trim()) continue;
+    let row;
+    try { row = JSON.parse(line); } catch { continue; }   // a torn line proves nothing; skip it
+    lines++;
+    if (row.k === "node" && row.id) nodes.set(row.id, { ...(nodes.get(row.id) || {}), ...row });
+    else if (row.k === "edge" && row.from && row.to && row.t) edges.set(`${row.from}|${row.t}|${row.to}`, row);
+  }
+  return { nodes, edges, lines };
+}
+
+/** Read a JSON file without throwing. @param {string} p Path. @returns {*} Parsed value or null. */
+function readJson(p) {
+  try { return JSON.parse(readFileSync(p, "utf8")); } catch { return null; }
+}
+
+/** Every `## Heading`-anchored id in a requirements or use-case document. */
+function idsIn(text, re) {
+  return [...new Set([...String(text || "").matchAll(re)].map((m) => m[1]))];
+}
+
+/**
+ * Project every artifact on disk for one run into nodes and edges.
+ *
+ * Pure: reads the tree, returns the graph it implies, writes nothing. That is what lets the same
+ * function serve both the incremental append and the backfill of a run recorded before this file
+ * existed — there is no second code path to keep in step.
+ *
+ * @param {string} cwd - Project root.
+ * @param {string} slug - Feature slug.
+ * @returns {{nodes: object[], edges: object[]}} Everything the artifacts imply, in a stable order.
+ */
+export function project(cwd, slug) {
+  const nodes = [], edges = [];
+  const node = (id, t, attrs = {}) => nodes.push({ k: "node", id, t, ...attrs });
+  const edge = (from, t, to) => edges.push({ k: "edge", from, t, to });
+
+  const root = localRoot(cwd, slug);
+  const receipt = readJson(receiptPath(cwd, slug));
+  const runId = runIdFromReceipt(receipt);
+  const runNode = runId ? `run:${runId}` : null;
+
+  // ---- work lineage --------------------------------------------------------------------------
+  if (runNode) {
+    node(runNode, "Run", {
+      run_id: runId, slug,
+      started_at: receipt?.started_at ?? null,
+      intake_sha256: receipt?.intake_sha256 ?? null,
+      auto_level: receipt?.config?.auto_level ?? null,
+    });
+  }
+
+  const oDir = ordersDir(cwd, slug);
+  const rDir = resultsDir(cwd, slug);
+  const orderByFile = new Map();
+  if (existsSync(oDir)) {
+    for (const f of readdirSync(oDir).filter((x) => x.endsWith(".json")).sort()) {
+      const o = readJson(join(oDir, f));
+      if (!o?.order_id) continue;
+      const id = `order:${o.order_id}`;
+      orderByFile.set(f, { id, order: o });
+      node(id, "Order", {
+        order_id: o.order_id, worker: o.worker ?? null, operation: o.operation ?? null,
+        round: o.round ?? null, attempt: o.attempt ?? null,
+        scope_id: o.payload?.scope_contract?.scope_id ?? o.payload?.scope_id ?? null,
+        run_id: o.run_id ?? runId ?? null,
+      });
+      if (runNode) edge(runNode, "PRODUCED", id);
+      const scopeId = o.payload?.scope_contract?.scope_id ?? o.payload?.scope_id ?? null;
+      if (scopeId) edge(id, "DERIVED_FROM", `scope:${slug}:${scopeId}`);
+    }
+  }
+
+  if (existsSync(rDir)) {
+    for (const f of readdirSync(rDir).filter((x) => x.endsWith(".json")).sort()) {
+      const r = readJson(join(rDir, f));
+      if (!r?.order_id) continue;
+      const id = `result:${r.order_id}`;
+      node(id, "Result", {
+        order_id: r.order_id, status: r.status ?? null,
+        artifacts: Array.isArray(r.artifacts) ? r.artifacts.length : 0,
+        discoveries: Array.isArray(r.discoveries) ? r.discoveries.length : 0,
+      });
+      edge(`order:${r.order_id}`, "PRODUCED", id);
+    }
+  }
+
+  // T0 verdicts — the artifact the evaluator is required to cite, and the reason the lineage half
+  // of this graph is worth having: a verdict node is the anchor of every "show me the evidence".
+  const vDir = verdictsDir(cwd, slug);
+  const verdictIdsByRound = new Map();     // round -> [verdict node id, ...], for the gate edge below
+  if (existsSync(vDir)) {
+    for (const f of readdirSync(vDir).filter((x) => x.endsWith(".json")).sort()) {
+      const v = readJson(join(vDir, f));
+      if (!v) continue;
+      const id = `verdict:${slug}:${basename(f, ".json")}`;
+      node(id, "Verdict", {
+        artifact: f, overall: v.overall ?? null, regression: !!v.regression,
+        round: v.round ?? null, attempt: v.attempt ?? null, scope_id: v.scope_id ?? null,
+        run_id: v.run_id ?? runId ?? null,
+      });
+      if (v.scope_id) edge(id, "EVALUATES", `scope:${slug}:${v.scope_id}`);
+      if (v.round != null && v.attempt != null && v.scope_id) {
+        edge(`order:${slug}/${v.scope_id}-r${v.round}-a${v.attempt}`, "PRODUCED", id);
+      }
+      if (v.round != null) {
+        if (!verdictIdsByRound.has(v.round)) verdictIdsByRound.set(v.round, []);
+        verdictIdsByRound.get(v.round).push(id);
+      }
+    }
+  }
+
+  // Gate crossings — `kernel/gate.mjs` is the sole writer of `gates.jsonl`; this reads the same
+  // ledger `verify/t0.mjs` writes for `trials.jsonl`, defensive per-line JSON.parse-or-skip.
+  //
+  // ID KEYED ON GATE ID + ORDINAL, not gate id alone, for the exact reason the trial key above is
+  // not the ordinal alone: L2 and L3 are crossed once per round, so a bare `gate:<slug>:L2` id would
+  // collapse every round's crossing onto one node — a repeat of the trial-key bug this file already
+  // learned from once. The ordinal counts a gate id's occurrences in ledger order, which also
+  // disambiguates round-independent gates (L1a, …) re-crossed across relaunches, where `round` alone
+  // is `null` on every row.
+  const gPath = gatesPath(cwd, slug);
+  if (existsSync(gPath)) {
+    const seenByGate = new Map();
+    for (const line of readFileSync(gPath, "utf8").split("\n")) {
+      if (!line.trim()) continue;
+      const g = readJson0(line);
+      if (!g?.gate) continue;
+      const n = (seenByGate.get(g.gate) || 0) + 1;
+      seenByGate.set(g.gate, n);
+      const id = `gate:${slug}:${g.gate}:${n}`;
+      node(id, "GateDecision", {
+        gate: g.gate, decision: g.decision ?? null, status: g.status ?? null,
+        source: g.source ?? null, note: g.note ?? null, round: g.round ?? null,
+        run_id: g.run_id ?? runId ?? null,
+      });
+      // A round-scoped gate's decision depends on that round's T0 verdict(s) — the most honest
+      // shape, since that is the evidence the decision was made against. Every other gate (and a
+      // round-scoped one with no verdict yet on disk) depends on the Run instead, so no
+      // `GateDecision` node is ever orphaned.
+      const roundVerdicts = (g.gate === "L2" || g.gate === "L3") ? verdictIdsByRound.get(g.round) : null;
+      if (roundVerdicts?.length) {
+        for (const vid of roundVerdicts) edge(id, "DEPENDS_ON", vid);
+      } else if (runNode) {
+        edge(id, "DEPENDS_ON", runNode);
+      }
+    }
+  }
+
+  // Trials — the ratchet's own history, with a real parent link already in the row.
+  const tPath = trialsPath(cwd, slug);
+  if (existsSync(tPath)) {
+    for (const line of readFileSync(tPath, "utf8").split("\n")) {
+      if (!line.trim()) continue;
+      const t = readJson0(line);
+      if (!t?.trial) continue;
+      // THE SCOPE IS PART OF THE KEY, because the ordinal alone is not unique. A trial ordinal
+      // counts within its scope, so `trial:<slug>:1` names one row per scope and every one of them
+      // collapsed onto a single node — the Map this graph is read back into keeps the last writer,
+      // so a scope's whole execution record vanished silently. Measured: four scopes' trials
+      // projected to two nodes. `--trace` is supposed to reach "the execution record"; it reached
+      // whichever scope happened to be written last. `baseline_trial` is chosen from the same
+      // scope's prior rows, so the SUPERSEDES edge resolves inside the same partition.
+      const trialKey = (n) => `trial:${slug}:${t.scope_id ? `${t.scope_id}:` : ""}${n}`;
+      const id = trialKey(t.trial);
+      node(id, "Trial", {
+        trial: t.trial, round: t.round ?? null, attempt: t.attempt ?? null,
+        scope_id: t.scope_id ?? null, status: t.status ?? null, artifact: t.artifact ?? null,
+        sha256: t.sha256 ?? null, run_id: t.run_id ?? runId ?? null,
+      });
+      if (t.baseline_trial) edge(id, "SUPERSEDES", trialKey(t.baseline_trial));
+      if (t.artifact) edge(id, "EVALUATES", `verdict:${slug}:${basename(String(t.artifact), ".json")}`);
+    }
+  }
+
+  // ---- domain --------------------------------------------------------------------------------
+  const sDir = scopesDir(cwd, slug);
+  if (existsSync(sDir)) {
+    for (const { contract } of readAllContracts(sDir, SCOPE_CONTRACT)) {
+      if (!contract?.scope_id) continue;
+      const id = `scope:${slug}:${contract.scope_id}`;
+      node(id, "Scope", {
+        scope_id: contract.scope_id, title: contract.title ?? null,
+        substrate: (contract.allowed_file_substrate || []).length,
+      });
+      // The domain family's entry edges. Before scope contracts anchored into the committed spec
+      // these could not be drawn: `covers` had this reader and no writer, and there was no
+      // Scope→UseCase relation at all, so a Scope sat in the graph as an isolated node beside the
+      // UseCase nodes it was built from and `--trace` stopped there instead of reaching the
+      // objective. The contract now names both, so both are projectable.
+      for (const uc of contract.use_cases || []) edge(id, "IMPLEMENTS", `uc:${slug}:${ucId(uc)}`);
+      for (const req of contract.covers || []) edge(id, "COVERS", `req:${slug}:${req}`);
+      for (const dep of contract.depends_on || []) edge(id, "DEPENDS_ON", `scope:${slug}:${dep}`);
+    }
+  }
+
+  const uDir = usecasesDir(cwd, slug);
+  if (existsSync(uDir)) {
+    for (const f of readdirSync(uDir).filter((x) => x.endsWith(".md") && x !== "_index.md").sort()) {
+      const ucId = basename(f, ".md");
+      node(`uc:${slug}:${ucId}`, "UseCase", { use_case: ucId, file: f });
+    }
+  }
+
+  const reqPath = requirementsPath(cwd, slug);
+  if (existsSync(reqPath)) {
+    for (const r of idsIn(readFileSync(reqPath, "utf8"), /^\s*[-*|]?\s*\**\s*(REQ-[A-Z0-9-]+)/gm)) {
+      node(`req:${slug}:${r}`, "Requirement", { req_id: r });
+    }
+  }
+
+  // The wiring map is literally a graph written as a table: per use case, engine → seam →
+  // entry-point call site → affordance. Reading it back as edges is the whole reason it is a table.
+  const wPath = wiringMapPath(cwd, slug);
+  if (existsSync(wPath)) {
+    // READ THE FIELDS THE SPEC ACTUALLY PRODUCES. This loop asked for `contract.wiring` while
+    // WIRING_MAP names the field `entries`, and for `row.seam`/`row.entry_point` while WiringEntry
+    // declares `wiring_seam`/`entry_call_site` — three name mismatches stacked on a parser that was
+    // returning nothing anyway. Every one of them independently yields an empty domain half, and
+    // none of them can fail loudly: an absent field and an absent wiring map look identical here.
+    // The legacy spellings stay as fallbacks so a map written either way still projects.
+    const found = readContract(wPath, WIRING_MAP);
+    for (const row of found?.contract?.entries || found?.contract?.wiring || found?.contract?.rows || []) {
+      const uc = row.use_case || row.uc;
+      const seam = row.wiring_seam || row.seam;
+      const entryPoint = row.entry_call_site ?? row.entry_point ?? null;
+      if (!uc || !seam) continue;
+      const seamId = `seam:${slug}:${seam}`;
+      node(seamId, "Seam", { seam, engine: row.engine ?? null, entry_point: entryPoint });
+      edge(`uc:${slug}:${ucId(uc)}`, "DEPENDS_ON", seamId);
+    }
+  }
+
+  return { nodes, edges };
+}
+
+/** JSON.parse for one line, without throwing. @param {string} l Line. @returns {*} Parsed or null. */
+function readJson0(l) {
+  try { return JSON.parse(l); } catch { return null; }
+}
+
+/**
+ * Append everything the artifacts imply that the graph does not already carry.
+ *
+ * @param {string} cwd - Project root.
+ * @param {string} slug - Feature slug.
+ * @returns {{path: string, nodes_added: number, edges_added: number, nodes_total: number, edges_total: number, backfilled: boolean}}
+ *   `backfilled` is true when the graph did not exist before this call — the migration case, which
+ *   is the same code path as the steady-state one.
+ */
+export function appendGraph(cwd, slug) {
+  const p = graphPath(cwd, slug);
+  const before = readGraph(cwd, slug);
+  const backfilled = before.lines === 0;
+  const { nodes, edges } = project(cwd, slug);
+
+  const out = [];
+  for (const n of nodes) {
+    const prev = before.nodes.get(n.id);
+    // Append only when something actually changed — the file is a log, and a log that repeats
+    // itself on every run stops being readable long before it stops being correct.
+    if (!prev || JSON.stringify({ ...prev, k: undefined }) !== JSON.stringify({ ...n, k: undefined })) out.push(n);
+  }
+  for (const e of edges) {
+    if (!before.edges.has(`${e.from}|${e.t}|${e.to}`)) out.push(e);
+  }
+
+  if (out.length) {
+    mkdirSync(dirname(p), { recursive: true });
+    appendFileSync(p, out.map((r) => JSON.stringify(r)).join("\n") + "\n");
+  }
+  const after = readGraph(cwd, slug);
+  return {
+    path: p,
+    nodes_added: out.filter((r) => r.k === "node").length,
+    edges_added: out.filter((r) => r.k === "edge").length,
+    nodes_total: after.nodes.size,
+    edges_total: after.edges.size,
+    backfilled,
+  };
+}
+
+/**
+ * The bounded subgraph a relaunch needs — the fast-forward as a query rather than a directory walk.
+ *
+ * @param {string} cwd - Project root.
+ * @param {string} slug - Feature slug.
+ * @returns {object} Counts and ids only. Deliberately small: the point of the graph is that the
+ *   orchestrator's working set is a subgraph, not a re-read of all state.
+ */
+export function runSubgraph(cwd, slug) {
+  const { nodes, edges, lines } = readGraph(cwd, slug);
+  const of = (t) => [...nodes.values()].filter((n) => n.t === t);
+  const orders = of("Order"), results = of("Result"), verdicts = of("Verdict");
+  const resultIds = new Set(results.map((r) => r.order_id));
+  const greenByRound = {};
+  for (const v of verdicts) {
+    if (v.overall !== "green" || v.round == null || !v.scope_id) continue;
+    (greenByRound[v.round] ||= new Set()).add(v.scope_id);
+  }
+  return {
+    graph_lines: lines,
+    run: of("Run")[0]?.run_id ?? null,
+    scopes: of("Scope").map((s) => s.scope_id).sort(),
+    use_cases: of("UseCase").map((u) => u.use_case).sort(),
+    requirements: of("Requirement").map((r) => r.req_id).sort(),
+    seams: of("Seam").map((s) => s.seam).sort(),
+    orders: orders.length,
+    pending_orders: orders.filter((o) => !resultIds.has(o.order_id)).map((o) => o.order_id).sort(),
+    rounds_with_green: Object.keys(greenByRound).map(Number).sort((a, b) => a - b),
+    green_scopes_by_round: Object.fromEntries(Object.entries(greenByRound).map(([r, s]) => [r, [...s].sort()])),
+    trials: of("Trial").length,
+    edges: edges.size,
+  };
+}
+
+/**
+ * Walk backwards from one node, following every edge that points at it or away from it.
+ *
+ * This is the reliability invariant made executable: pick a verdict, walk to the order that
+ * produced it, the scope it evaluates, the requirements that scope covers, and the trials that
+ * superseded it — in one query, over one file, instead of a grep across four.
+ *
+ * @param {string} cwd - Project root.
+ * @param {string} slug - Feature slug.
+ * @param {string} startId - The node to trace from.
+ * @param {number} [depth=4] - How many hops to follow.
+ * @returns {{start: string, found: boolean, path: object[]}} Each hop as `{from, t, to, node}`. A
+ *   hop whose far node carries `missing: true` is a dangling edge — the artifact it names is not on
+ *   disk. That is reported rather than dropped: an edge to nothing is a fact about the run.
+ */
+export function trace(cwd, slug, startId, depth = 4) {
+  const { nodes, edges } = readGraph(cwd, slug);
+  if (!nodes.has(startId)) return { start: startId, found: false, path: [] };
+  const seen = new Set([startId]);
+  let frontier = [startId];
+  const path = [];
+  for (let d = 0; d < depth && frontier.length; d++) {
+    const next = [];
+    for (const e of edges.values()) {
+      for (const [near, far] of [[e.from, e.to], [e.to, e.from]]) {
+        if (!frontier.includes(near) || seen.has(far)) continue;
+        seen.add(far);
+        next.push(far);
+        path.push({ from: e.from, t: e.t, to: e.to, node: nodes.get(far) ?? { id: far, t: "?", missing: true } });
+      }
+    }
+    frontier = next;
+  }
+  return { start: startId, found: true, path };
+}
+
+export const ARGV_SPEC = {
+  usage: "harness.mjs reduce graph --slug <slug> [--cwd <dir>] [--subgraph run] [--trace <node-id>] [--depth N]",
+  _: { arity: 0, max: 0, name: "(no positional operands)" },
+  slug: { type: "str", required: true },
+  cwd: { type: "path" },
+  subgraph: { type: "enum", values: ["run"] },
+  trace: { type: "str" },
+  depth: { type: "int", min: 1, max: 12, default: 4 },
+};
+
+/**
+ * Append everything the artifacts imply, then optionally answer a query over the result.
+ *
+ * @param {string[]} rawArgv - The subcommand's own arguments (harness.mjs strips the verb words).
+ * @returns {void} Exits 0; prints the append report, or the requested query, as JSON.
+ */
+export function cli(rawArgv) {
+  const args = runArgs(ARGV_SPEC, rawArgv);
+  const cwd = resolve(args.cwd || process.cwd());
+  const report = appendGraph(cwd, args.slug);
+  if (args.trace) { console.log(JSON.stringify(trace(cwd, args.slug, args.trace, args.depth), null, 2)); process.exit(0); }
+  if (args.subgraph) { console.log(JSON.stringify(runSubgraph(cwd, args.slug), null, 2)); process.exit(0); }
+  console.log(JSON.stringify(report, null, 2));
+  process.exit(0);
+}

package/kernel/reduce/hill.mjs

@@ -0,0 +1,152 @@
+#!/usr/bin/env node
+// Hill Phase Derivation — phases are derived from facts, never authored
+// Mechanical derivation of a scope's hill position based on facts (T0 verdicts, T1 evaluation, and discovery ledger).
+// Writes to shapeup/<slug>/hill/<scope-id>.yml
+
+import { readFileSync, writeFileSync, existsSync, readdirSync, mkdirSync } from "node:fs";
+import { resolve, join } from "node:path";
+import { runArgs } from "../lib/argv.mjs";
+import { scopesDir, hillDir, verdictsDir, evaluationDir, discoveryLedger } from "../lib/paths.mjs";
+import { readAllContracts, SCOPE_CONTRACT } from "../lib/contract.mjs";
+
+/**
+ * Derive and write the hill phase for all scopes mechanically based on T0, T1, and ledger facts.
+ * 
+ * The derived phase follows these progression rules (facts move dots, not authors):
+ * - UPHILL_UNKNOWN: open unknowns > 0 in the ledger for this scope
+ * - UPHILL_SOLVED: unknowns 0, no T0-green yet
+ * - DOWNHILL_EXECUTION: ≥1 T0-green; T1/seesaw pending
+ * - FINISHED: T1 PASS ∧ seesaw green
+ *
+ * @param {string} cwd - The project root directory.
+ * @param {string} slug - The feature slug being built.
+ * @returns {Array<{scope_id: string, phase: string, changed: boolean}>} A report of all scopes processed, their derived phase, and whether the hill shard on disk was modified.
+ *   Side effects: writes to `shapeup/<slug>/hill/<scope-id>.yml` for each scope.
+ */
+export function deriveHill(cwd, slug) {
+  const scopes = readAllContracts(scopesDir(cwd, slug), SCOPE_CONTRACT).map((x) => x.contract);
+  const vDir = verdictsDir(cwd, slug);
+  const evalDir = evaluationDir(cwd, slug);
+  const ledgerPath = discoveryLedger(cwd, slug);
+  const hDir = hillDir(cwd, slug);
+  
+  if (!existsSync(hDir)) mkdirSync(hDir, { recursive: true });
+  
+  // 1. Check if T1 Evaluation passed (spec-conformance === PASS for the most recent run)
+  let t1Pass = false;
+  const evalFile = join(evalDir, ".verdicts-run.jsonl");
+  if (existsSync(evalFile)) {
+    const lines = readFileSync(evalFile, "utf8").trim().split(/\n/).filter(Boolean);
+    let maxRun = 0;
+    for (const line of lines) {
+      try {
+        const parsed = JSON.parse(line);
+        if (parsed.run >= maxRun) {
+          maxRun = parsed.run;
+          if (parsed.dimension === "spec-conformance") {
+            t1Pass = (parsed.verdict === "PASS");
+          }
+        }
+      } catch (e) {
+        // ignore parse errors
+      }
+    }
+  }
+
+  // 2. T0 facts per scope: has it achieved a green overall verdict? was seesaw also green?
+  const t0Facts = {};
+  if (existsSync(vDir)) {
+    for (const f of readdirSync(vDir)) {
+      if (!f.endsWith(".json")) continue;
+      try {
+        const b = JSON.parse(readFileSync(join(vDir, f), "utf8"));
+        if (!t0Facts[b.scope_id]) t0Facts[b.scope_id] = { hasGreen: false, seesawGreen: false };
+        if (b.overall === "green") {
+          t0Facts[b.scope_id].hasGreen = true;
+          // Read the REAL seesaw result off the verdict artifact (`t0.mjs`'s `writeArtifact()`
+          // already persists the full `{ran, pass, scopes_checked, failing}` object), rather than
+          // inferring it from a false `regression` flag. That inference was vacuously true on every
+          // green T0 whether or not a seesaw check ever ran: nothing in this codebase currently
+          // passes `--seesaw-registry` to `verify t0`, so `seesaw.ran` is always `false` today and
+          // `regression` is always `false` too — "not asked" was being read as "clean," letting a
+          // scope reach FINISHED on a regression check that had never executed.
+          //
+          // Betting Table decision (Phase 3.5 / S4): wiring the seesaw registry for real is a
+          // genuine feature with a real running cost (re-running every finished scope's fixtures
+          // on every later attempt) and is out of proportion to a certification-gap fix. Deferred,
+          // not silently dropped — a scope with no registry wired simply cannot reach FINISHED via
+          // this path today, which is the honest state of the system: this check was never really
+          // gating FINISHED before either.
+          if (b.seesaw?.ran && b.seesaw?.pass) {
+            t0Facts[b.scope_id].seesawGreen = true;
+          }
+        }
+      } catch (e) {
+        // ignore parse errors
+      }
+    }
+  }
+  
+  // 3. Ledger unknowns per scope
+  const scopeUnknowns = {};
+  if (existsSync(ledgerPath)) {
+    const lines = readFileSync(ledgerPath, "utf8").split("\n");
+    let currentScope = null;
+    for (const line of lines) {
+      const m = line.match(/^## Discovered — .*?:([\w.-]+)-a\d+/);
+      if (m) {
+        currentScope = m[1];
+      }
+      if (currentScope && line.startsWith("~ ")) {
+        scopeUnknowns[currentScope] = (scopeUnknowns[currentScope] || 0) + 1;
+      }
+    }
+  }
+  
+  const report = [];
+  for (const s of scopes) {
+    const id = s.scope_id;
+    const t0 = t0Facts[id] || { hasGreen: false, seesawGreen: false };
+    const unknowns = scopeUnknowns[id] || 0;
+    
+    let phase = "UPHILL_UNKNOWN";
+    if (t1Pass && t0.hasGreen && t0.seesawGreen) {
+      phase = "FINISHED";
+    } else if (t0.hasGreen) {
+      phase = "DOWNHILL_EXECUTION";
+    } else if (unknowns === 0) {
+      phase = "UPHILL_SOLVED";
+    }
+    
+    const yaml = `scope_id: ${id}\nphase: ${phase}\n`;
+    const out = join(hDir, `${id}.yml`);
+    let changed = false;
+    if (!existsSync(out) || readFileSync(out, "utf8") !== yaml) {
+      writeFileSync(out, yaml);
+      changed = true;
+    }
+    report.push({ scope_id: id, phase, changed });
+  }
+  return report;
+}
+
+export const ARGV_SPEC = {
+  usage: "harness.mjs reduce hill --slug <slug> [--cwd <dir>]",
+  _: { arity: 0, max: 0, name: "(no positional operands)" },
+  slug: { type: "str", required: true },
+  cwd: { type: "path" },
+};
+
+/**
+ * Derive each scope's hill phase from its T0 and evaluation artifacts.
+ *
+ * @param {string[]} rawArgv - The subcommand's own arguments (harness.mjs strips the verb words).
+ * @returns {(Promise<void>|void)} Settles when the subcommand has written its output; most paths
+ *   call `process.exit()` with the subcommand's documented code rather than returning.
+ */
+export async function cli(rawArgv) {
+  const args = runArgs(ARGV_SPEC, rawArgv);
+  const cwd = resolve(args.cwd || process.cwd());
+  const report = deriveHill(cwd, args.slug);
+  console.log(JSON.stringify(report, null, 2));
+}